{"id":"30139b6b-1cac-4541-8f6c-b647a62ba9b6","arxiv_id":"2412.19778","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The paper proposes that Calabi-Yau flat metrics inherit extra symmetries from the surrounding space, and uses this to build compact symbolic approximations and exact expressions on special loci.","lead":"This paper studies the shape of the elusive Ricci-flat metric on symmetric Calabi-Yau spaces, the extra dimensions of string theory. It finds that the metric's correction term may have hidden extra symmetries, which lets the authors write short formulas that approximate it accurately.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Conjecture 3.2 is the load-bearing premise; the paper's salience and restricted-feature evidence is suggestive but lacks a direct phase-invariance test on a high-accuracy unconstrained model, leaving the exact equimodular metric formula unverified.","rationale":"The reader's conditional verdict is well aligned with the paper's own presentation: the authors label Conjecture 3.2 as unproved, and the Dwork and Cefalu examples show the symmetry mechanism is not universal. I do not find an internal inconsistency in the conditional arguments; Proposition 3.3 does follow from the conjecture because two points of X with equal |Z_i| are related by a U(1)^n transformation that preserves the ambient extension. The weakness is epistemic: the exact structural claim is supported mainly by approximate restricted-feature training. My proposed check would turn this into a direct falsifiable prediction. If it passes, the conditional theorems are on much firmer ground; if it fails, the verdict should move toward rejection of the central claim. The paper also contains independent contributions, notably the proven integration-weights identity (Theorem 3.4) and the explicit torus potential (Proposition 5.1). Since the paper is explicit about the conditional status and the numerical contribution is genuine, maintaining the conditional verdict is appropriate.","tokens_in":29511,"tokens_out":12088,"duration_ms":137703,"concrete_test":"Train an unconstrained spectral network (no phase symmetry built in) on the Fermat quintic to the lowest sigma achievable, or run Donaldson's balanced metric at the highest feasible degree k. Draw 10^4 pairs (p,q) of points on X with identical |Z_i| for all i but phases chosen so that q = (e^{i theta_i} Z_i) lies on X. Compute delta = |phi(p) - phi(q)| / std(phi). If max delta is not at the numerical-noise level (say > 10^-3), Conjecture 3.2 is falsified for the learned metric; if delta stays at machine precision over all pairs, the phase-independence claim survives a genuinely non-circular test. Optionally, on the same equimodular points, compare the eigenvalue ratio g^flat / iota^* g_FS to (n+1) pi lambda as a direct check of Prop. 6.5.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Conjecture 3.2 is the load-bearing premise: every nontrivial theorem in the paper—Prop. 3.3 (|Z_i|-only dependence of phi), Prop. 4.1 (ModNet feature set), and Prop. 6.5 (exact flat metric on the equimodular locus)—is conditional on it, and the paper proves no case of it. The empirical support is indirect: salience plots (Figs. 5, 7, 8) show phase-dependent features at the level of random noise, and ModNet built only from |Z_i|-dependent symmetric polynomials reaches low sigma-loss. But salience is a property of one trained network, not of the true flat potential, and low sigma-loss does not rigorously certify convergence to the unique flat metric (Section 2.3 and the Appendix C footnote explicitly note that sigma is a surrogate and can be small for pathological metrics). Thus the observed numerics are consistent with approximate phase-independence but do not establish the exact symmetry that the propositions require. If Conjecture 3.2 fails, the equimodular formula loses its foundation and the symbolic distillations become curve fits rather than consequences of a symmetry principle.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies Ricci-flat Kähler metrics on Fermat Calabi–Yau hypersurfaces. It introduces Conjecture 3.2 ('extrinsic symmetries'), which asserts the existence of an ambient extension of the Kähler potential whose symmetry group equals that of ||∇Q||. Under this conjecture, the paper proves that φ depends only on coordinate moduli (Props. 3.3 and 4.1), derives an exact proportionality g♭ = (n+1)πλ ι*g_FS on the equimodular locus (Prop. 6.5), and gives a pseudo-origin formula (Prop. 6.4). The paper also presents salience analyses of neural-network approximations, a new model (ModNet) built from symmetric polynomials in |Z_i|, and symbolic distillations of φ, including a compressed form with exponent 1/π. All main theoretical consequences are explicitly conditional on Conjecture 3.2, which is not proved in any special case.","tokens_in":29864,"tokens_out":15478,"duration_ms":157846,"significance":"The cleanest rigorous result is Theorem 3.4, an analytic identity for the integration weights; its proof in Appendix A is complete and checkable. If Conjecture 3.2 holds, the equimodular formula of Prop. 6.5 is a rare exact statement about a Calabi–Yau Ricci-flat metric, and the feature reduction behind ModNet is well motivated. The authors are transparent about the surrogate nature of the σ-loss (Sec. 2.3, App. C). However, the central conjecture is unproved, and the empirical support is indirect: salience measures properties of one trained network, and low σ-loss does not certify exact phase invariance. The paper is honest about these limitations, and the computational improvements are suggestive, but the significance of the theoretical claims is conditional.","major_comments":[{"comment":"Conjecture 3.2 is the load-bearing premise for the main theoretical results, and no nontrivial case of the conjecture is proved. The supporting evidence is indirect: salience plots (Figs. 5, 7, 8) and low σ-loss of ModNet (Table 1) are consistent with approximate phase-independence, but salience is a property of one trained model and the σ-measure is a surrogate loss (Sec. 2.3). A direct test would substantially strengthen the paper: train an unconstrained high-capacity model with phase-dependent features on the Fermat quintic, and report the maximum deviation |φ(Z)−φ(e^{iθ}Z)| over the manifold as a function of training loss and network capacity. Without such a test, Proposition 6.5 remains a consequence of an unverified conjecture.","section":"Sec. 3.3, Props. 3.3, 4.1, 6.5"},{"comment":"The counterexample in the proof of Proposition 4.2 is invalid as written. The locus is stated to satisfy Σ Z_k^n = 0, but the Fermat hypersurface is defined by Σ Z_k^{n+1} = 0 (Eqs. 2.1 and 2.2). Moreover, the test function cos(n arg Z_1) is invariant only under n-th root phase rotations of Z_1, not under the full toric symmetry group Z_{n+1}^{n-1} of the Fermat hypersurface. Therefore the construction does not demonstrate that a generic Isom(CY)-invariant 0-form can depend on phases. Since the proposition is used to argue that |Z_i|-dependence is a novel consequence of Conjecture 3.2, the proof must be corrected or replaced.","section":"Appendix A.2, Prop. 4.2"},{"comment":"The numerical claims, including the assertion that ModNet 'beats previous ML approaches with respect to every metric' (Table 2), are reported without error bars, seeds, or code. Several comparison entries in Table 1 are explicitly inferred from figures in prior work. For a paper in which numerical evidence supports the central conjecture, the authors should report standard deviations over multiple training seeds and specify all hyperparameters, data splits, and hardware/software versions. This is needed to assess whether the improvements over prior models are significant.","section":"Tables 1 and 2, Sec. 4.1.1"},{"comment":"The proof of Proposition 6.5 is only a sketch. The key step — that, under Conjecture 3.2, the chain rule and permutation symmetry imply ∂∂φ|_X = aI + bZ†⊗Z on the equimodular locus — is not shown, and the derivation of the coefficient (n+1)πλ from det g♭ = κΩ∧Ω̄ is not displayed. Since this is one of the central exact results, the proof should be expanded to the same level of detail as Lemma A.3 so that the constant is verifiable.","section":"Prop. 6.5"}],"minor_comments":[{"comment":"The denominator in the σ-measure is written as κΩΩ; this should be κ Ω∧Ω̄ (or the text should clarify the notation for the volume form).","section":"Eq. (2.6)"},{"comment":"The notation Z^{n-1}_n for the toric symmetry should read Z_{n+1}^{n-1} for the Fermat hypersurface; as printed it is inconsistent with the degree of the defining polynomial.","section":"Appendix A.2"},{"comment":"The remark that ∂iφ/Zi and ∂i∂jφ/(ZiZj) are real under Conjecture 3.2 appears to be false: if φ depends only on |Z_i|, the quantity ∂iφ/Zi acquires a phase factor ar Z_i/Z_i under independent U(1) rotations, so it is not invariant. The inequalities need to be reformulated, for example in terms of moduli of the derivatives.","section":"Conjecture 6.3"},{"comment":"The exponent 1/π in Eq. (5.2) is described as determined empirically; the authors should state explicitly that this is a fit parameter and that the quoted σ-loss corresponds to the fitted value, not to a derived constant.","section":"Sec. 5.1"}],"recommendation":"major_revision","confidential_remarks":"The paper has a clean core result (Theorem 3.4) and an honest, well-written treatment of its main conjecture. The major obstacle is that the central theoretical claims are conditional on an unproved conjecture, and the empirical evidence, while suggestive, does not currently include a direct test of exact phase invariance. The proof of Proposition 4.2 also contains a clear degree-n vs degree-(n+1) slip that should be corrected. With a direct numerical test, corrected proofs, and reproducibility details for the ML experiments, the paper could be suitable for publication. The manuscript is somewhat overlong for the amount of rigorous content, but the breadth of exploratory results is appropriate for a hep-th audience."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read of Mirjanić and Mishra. The headline is: this is a genuinely new idea — that the Ricci-flat potential on Fermat CYs may admit an ambient extension with more symmetry than the manifold itself — but the central claim is a conjecture, and the evidence, while suggestive, does not close the case. The paper is honest about that; Conjecture 3.2 is clearly labeled and all strong consequences are derived conditionally.\n\nWhat is actually new and good: the integration weights identity (Theorem 3.4, w = ∥Z∥^{2n}/∥∇Q∥^2) is a clean, proven statement with an elegant proof. Proposition 4.2 properly shows that |Zi|-dependence of φ is not implied by the isometry group for d > 1, so the conjecture does real theoretical work. The symbolic distillation results — compact closed forms like Eq. (5.2) — are interesting, and the σ-losses are competitive. The practical message that a model built only from |Zi|-symmetric features can match or beat more general networks is useful.\n\nThe soft spots, in proportion. First, Conjecture 3.2 is load-bearing and unproved. Every strong consequence — the phase-invariance of φ, the equimodular locus formula (Prop 6.5) — rests on it. The proof of 6.5 is really a derivation under the conjecture, not an independent verification. Second, the empirical support is indirect. Salience plots show phase-dependent features at noise level, and ModNet performs well, but that does not establish exact phase-invariance. Low σ-loss does not certify convergence to the true flat metric, a point the authors themselves make in Sections 2.3 and Appendix C. So the exact symmetry remains a conjecture. Third, the numerical results lack code, error bars, and repeated-run statistics, and some comparison numbers are inferred from figures. That matters for a paper whose main evidence is numerical. Fourth, there is an element of circularity: the conjecture selects the feature set, then the model's performance is cited as evidence for the conjecture. The authors seem aware of this, and the Dwork non-example is a good faith attempt to bound the claim.\n\nWho is this for? People working on numerical CY metrics, especially ML-based approaches, and string theorists who need flat metric approximations. It deserves a serious referee. I would send it out and ask for code, seed-averaged statistics, and a non-circular test — for example, predicting φ on a deformed family not used for feature selection. If those checks come back clean, this could be a useful contribution.","headline":"A genuinely new symmetry idea for Fermat CY metrics, but the central claim is an unproved conjecture and the numerical evidence is suggestive rather than conclusive.","tokens_in":30325,"tokens_out":2802,"would_cite":true,"duration_ms":27782,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14J32","32Q25","53C55","68T07"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that Ricci-flat Kähler potentials on Fermat Calabi–Yau hypersurfaces are governed by extrinsic symmetries of the ambient space, so the potential depends only on coordinate absolute values and the flat metric is fixed…","keywords":["Calabi-Yau manifolds","Ricci-flat metrics","extrinsic symmetries","Kähler potential","Monge-Ampère equation","Fermat hypersurfaces","neural network approximations","symbolic regression"],"falsifier":"On the Fermat quintic, fix all $|Z_i|$ and all phases except $\\arg Z_1$, and compute the Monge–Ampère-optimized $\\phi$ while rotating $\\arg Z_1$; a measurable variation of $\\phi$ would falsify the phase-independence predicted by Conjecture 3.2. A second check is to construct an explicit ambient extension $\\phi_P$ and compare $\\mathrm{Sym}(\\phi_P)$ with $\\mathrm{Sym}(\\|\\nabla Q\\|)$ directly.","tokens_in":29319,"feed_emoji":"📐","tokens_out":9238,"duration_ms":91642,"temperature":0.7,"pith_summary":"This paper tries to turn Yau's non-constructive existence theorem for Ricci-flat metrics on Calabi–Yau manifolds into explicit analytic and symbolic control on the Fermat family. The central conjecture is that the Kähler potential $\\phi$ extends from the hypersurface to the whole projective space with exactly the symmetries of $\\|\\nabla Q\\|$, the gradient length of the defining polynomial. If this conjecture holds, a coordinate appearing only as $Z_i^{n+1}$ in $Q$ enters $\\phi$ only through $|Z_i|$, and on the equimodular locus the flat metric is exactly $(n+1)\\pi\\lambda\\,\\iota^*g_{FS}$. This matters because it would give physicists and geometers concrete, checkable expressions for a metric that has been known to exist for half a century but never explicitly constructed.","feed_headline":"Ricci-flat metrics on Fermat Calabi-Yaus shrink to few parameters","feed_subtitle":"A conjectured ambient symmetry forces Kähler potentials to ignore phases and fixes the metric exactly on a key locus.","key_machinery":"The load-bearing object is the extrinsic symmetry group $\\mathrm{Sym}(\\|\\nabla Q\\|)$ of the ambient gradient-length function, together with the conjecture that the Kähler potential extends to an ambient function $\\phi_P$ with exactly that symmetry group. Because $\\|\\nabla Q\\|$ forgets phases of coordinates that occur only as pure powers, this group can be continuous even when the Calabi–Yau itself has only discrete isometries. The machinery converts metric information into a finite set of invariant features: for Fermat hypersurfaces these are the normalised power sums $s_k=\\sum_i|Z_i|^{2k}/(\\sum_i|Z_i|^2)^k$, which the paper feeds to a neural network called ModNet and to symbolic-regression distillation. The same ambient extension also lifts the integration weights through the identity $w=\\|Z\\|^{2n}/\\|\\nabla Q\\|^2$, connecting the Monge–Ampère normalisation of the flat metric to the gradient norm that defines the symmetry.","core_discovery":"Working in the PhiModel representation $g^\\flat=\\iota^*(g_{FS}+\\partial\\bar\\partial\\phi)$, the paper's central claim is that the correction scalar $\\phi$ carries more symmetry than the Calabi–Yau manifold itself. Conjecture 3.2 states that for a hypersurface defined by $Q=0$ there is an ambient function $\\phi_P$ agreeing with $\\phi$ on the hypersurface and satisfying $\\mathrm{Sym}(\\phi_P)=\\mathrm{Sym}(\\|\\nabla Q\\|)$, a group that can be strictly larger than the isometry group of the Calabi–Yau. From this, Proposition 3.3 derives a coordinate-wise $U(1)$ invariance: whenever $Z_i$ appears in $Q$ only as $Z_i^{n+1}$, the potential is independent of $\\arg Z_i$ and depends on $|Z_i|$ alone; Proposition 4.1 upgrades this to the full Fermat family. On the equimodular locus, Proposition 6.5 turns the same symmetry into an exact identity $g^\\flat|_X=(n+1)\\pi\\lambda\\,\\iota^*g_{FS}$, where $\\lambda$ is fixed by the volume normalisation, so a positive-codimension sublocus of the flat metric is known analytically. The paper argues that these consequences are strongly supported by neural-network salience experiments and by distilled closed-form approximations that keep the same loss with far fewer parameters.","pith_inferences":["An extension the authors leave implicit is that the same phase-independence should hold on any hypersurface whose defining polynomial is a sum of pure powers, so the results would transfer to non-Fermat but monomial-defined Calabi–Yau hypersurfaces.","The empirical relation $\\phi^{-1}\\sim\\|\\nabla\\phi\\|_1$ points toward a first-order PDE distinct from Monge–Ampère; proving such a PDE could provide an independent route to Conjecture 3.2 and explain the recurring corner structure in the plots.","The cross-dimensional self-similarity suggests a dimension-independent master potential; if real, it would allow formulas derived on the torus to be lifted directly to higher-dimensional Fermat manifolds.","The observed near-zero upper range of the third Chern form suggests a pointwise curvature bound $c_3\\le 0$ for the flat metric; confirming this numerically on the sextic would test the same structural hypothesis in a new dimension."],"forward_implications":["On every Fermat Calabi–Yau, the correction $\\phi$ is a function of $|Z_1|,\\dots,|Z_n|$ alone, so future metric computations can discard all phase information in the coordinates without losing expressive power.","The flat metric on the equimodular locus is known in closed form, $g^\\flat|_X=(n+1)\\pi\\lambda\\,\\iota^*g_{FS}$, giving an analytic target that numerical approximations can be tested against.","Distilled symbolic formulas such as $\\mathrm{poly}(s_i)^{1/\\pi}$ match the $\\sigma$-loss of large neural networks, about $10^{-3}$ on the quintic, with only a handful of fitted parameters.","The integration-weight identity $w=\\|Z\\|^{2n}/\\|\\nabla Q\\|^2$ makes the global 0-form $w$ computable from ambient data, simplifying curvature integrals and Monte-Carlo sampling.","For the 0–1 mixed family, the symmetry argument predicts that only the $Z_0$–$Z_1$ mixed feature survives, and the paper's salience experiments confirm that prediction."],"supporting_citations":[{"why":"Yau's theorem supplies the unique Ricci-flat metric whose non-constructive nature motivates the whole investigation.","marker":"[1]"},{"why":"Donaldson's balanced-embedding algorithm is the classical numerical benchmark for approximate Calabi–Yau metrics.","marker":"[2]"},{"why":"Donaldson's numerical results provide the comparison point for the size and complexity of the metric ansatz.","marker":"[3]"},{"why":"Headrick and Nassar's energy functional is the state-of-the-art non-machine-learning baseline quoted throughout the paper.","marker":"[4]"},{"why":"Jejjala et al. established neural-network approximations to Calabi–Yau metrics that this paper extends with symmetry arguments.","marker":"[21]"},{"why":"Larfors et al. introduced the PhiModel representation $g=\\iota^*(g_{FS}+\\partial\\bar\\partial\\phi)$ that the paper uses for the potential.","marker":"[35]"},{"why":"Berglund et al. provided the spectral network baseline and the training setup whose salience and loss are compared with ModNet.","marker":"[37]"},{"why":"This reference supplies the machine-learning training pipeline used for the numerical experiments.","marker":"[46]"}],"fun_headline_variants":["Hidden symmetries flatten Calabi-Yau Ricci-flat metrics","Symmetries beyond manifold exact Ricci-flat on a locus","Fermat CY symmetries force compact metric representation","Ambient symmetry yields closed-form Calabi-Yau metric","Extrinsic symmetries shrink neural Ricci-flat models"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything hinges on Conjecture 3.2: that the Kähler potential can be extended off the Calabi–Yau to the ambient projective space so that it has exactly the symmetries of the gradient-length function of the defining polynomial; if no such extension exists, the phase-independence reduction and the exact locus formulas lose their foundation.","fun_headline_variants_meta":{"raw":{"variants":["Hidden symmetries flatten Calabi-Yau Ricci-flat metrics","Symmetries beyond manifold exact Ricci-flat on a locus","Fermat CY symmetries force compact metric representation","Ambient symmetry yields closed-form Calabi-Yau metric","Extrinsic symmetries shrink neural Ricci-flat models"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000241,"raw_usage":{"total_tokens":1578,"prompt_tokens":1061,"completion_tokens":517,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":677,"completion_tokens_details":{"reasoning_tokens":437}},"tokens_in":677,"tokens_out":517,"duration_ms":6053,"temperature":1.0,"reasoning_tokens":437,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T23:51:36.599866+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On the Fermat quintic, fix all $|Z_i|$ and all phases except $\\arg Z_1$, and compute the Monge–Ampère-optimized $\\phi$ while rotating $\\arg Z_1$; a measurable variation of $\\phi$ would falsify the phase-independence predicted by Conjecture 3.2. A second check is to construct an explicit ambient extension $\\phi_P$ and compare $\\mathrm{Sym}(\\phi_P)$ with $\\mathrm{Sym}(\\|\\nabla Q\\|)$ directly.","supporting_citations":[{"cited_title":"Headrick and A","cited_arxiv_id":null,"evidence_quote":"Headrick and Nassar's energy functional is the state-of-the-art non-machine-learning baseline quoted throughout the paper."},{"cited_title":"Larfors, A","cited_arxiv_id":null,"evidence_quote":"Larfors et al. introduced the PhiModel representation $g=\\iota^*(g_{FS}+\\partial\\bar\\partial\\phi)$ that the paper uses for the potential."}],"review_version":1}